Hi @niccolocappelli.business,
Sorry you’re running into this on first boot.
Based on the current public System Recovery guide, the documented recovery path is still to use a keyboard and display connected directly to the DGX Spark, power it on, and immediately hold Esc or Del to enter UEFI. From there, the guide says to restore UEFI defaults, confirm Secure Boot is enabled / restore factory keys, and then use Save & Exit -> Boot Override to boot the recovery USB.
A few direct answers to your questions:
-
I’m not seeing a separate public “forced recovery” or hardware reset sequence in the DGX Spark docs beyond the UEFI + recovery USB process above. The guide specifically calls out using a keyboard plugged directly into the Spark, and notes that some Bluetooth keyboards may not work during this process.
-
Yes, interrupted first-boot updates are a plausible cause here. The Initial Setup - First Boot page explicitly warns not to shut down or reboot during the initial update process and says powering down during updates can cause system damage.
-
I’m not seeing a separate public hardware-level reset procedure in the user guide. One additional sanity check: please make sure you are using the supplied power adapter, since the Known Issues page notes that other adapters may reduce performance, prevent boot, or cause unexpected shutdowns.
-
If the system still cannot enter UEFI at all after:
-
minimal peripherals only
-
a plain wired USB keyboard connected directly to the unit
-
the supplied power adapter
-
and a correctly created recovery USB
then this is beyond the documented self-recovery flow, since the published recovery steps depend on entering UEFI and selecting the USB from Boot Override.
If you are able to get the system booted far enough to run diagnostics, please run the DGX Spark field diagnostic and post the results here.
Field diagnostic install / run steps (public sources for review: Contacting NVIDIA Support, NVIDIA DGX SPARK Field Diagnostics):
sudo mkdir -p /usr/share/keyrings
curl -fsSL https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/cuda-archive-keyring.gpg | sudo tee /usr/share/keyrings/cuda-archive-keyring.gpg > /dev/null
echo "deb [signed-by=/usr/share/keyrings/cuda-archive-keyring.gpg] https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa /" | sudo tee /etc/apt/sources.list.d/cuda-sbsa-ubuntu2404.list
sudo apt-get update
sudo apt-get install dgx-spark-fieldiag
dpkg -l | grep dgx-spark-fieldiag
Before running it, disable Secure Boot:
sudo mokutil --sb-state
sudo systemctl reboot --firmware-setup
Then in UEFI go to Security -> Secure Boot -> Disable Secure Boot, save changes, and reboot.
Run the diagnostic:
sudo init 3
cd /opt/nvidia/dgx-spark-fieldiag
sudo ./partnerdiag --field
This switches the display to TTY/text console mode, which is expected. The run takes about 30 minutes and ends with a PASS / FAIL / RETEST banner. If the run is interrupted, power-cycle the system before trying again.
After the run, re-enable Secure Boot:
sudo systemctl reboot --firmware-setup
Then in UEFI go to Security -> Secure Boot -> Enable Secure Boot, save, and reboot.
Please post back with:
-
the final PASS / FAIL / RETEST banner
-
the contents of the latest summary.json
-
and ideally a compressed copy of the latest log directory
The field diagnostic generates logs in the dgx/logs-<yyyymmdd-hhnnss> folder. That folder includes summary.json, plus general run logs such as run.log and output.log.
If you cannot get into UEFI at all, then field diagnostics will likely have to wait until we can first get the system to boot recovery media or otherwise recover the platform.